How I Actually Use AI in the Design Process

Every few weeks someone asks me whether AI is going to replace designers. My honest answer: no, but it's already changed how I work in ways I didn't expect. Here's a concrete look at what my AI-assisted workflow actually looks like — no hype, no doom, just the practical reality.

Where AI actually helps

I use AI tools daily, but not in the ways most people assume. I'm not generating layouts or asking ChatGPT to design logos. The value shows up in less glamorous places:

Code generation and scaffolding

This is the biggest win. I use Claude as a pair programmer for about 60% of my development work. It's exceptional at generating boilerplate — API routes, database schemas, component scaffolding, TypeScript types. Things that are tedious but critical. What used to take an afternoon now takes twenty minutes.

The key is specificity. Vague prompts get vague code. I write detailed specifications — what the function should accept, what it should return, what edge cases to handle — and the output is genuinely useful. Treat it like a junior developer who's very fast but has no taste.

Content iteration

For marketing copy, blog posts, and documentation, AI is an excellent first-draft machine. I'll write a rough outline, expand it with AI, then edit heavily. The editing step is non-negotiable. AI prose is fluent but generic — it needs a human pass to add voice, remove clichés, and cut the filler.

Research and synthesis

When I'm exploring a new domain — music rights law, for example — AI is invaluable for getting up to speed quickly. It won't replace reading primary sources, but it's a fast way to build a mental model before diving deeper.

Where AI falls short

There are areas where I've tried AI tools and consistently gone back to manual work:

• Visual design decisions. AI can generate images, but it can't make taste-driven choices about hierarchy, whitespace, or typographic pairings. Design is judgment, not generation.

• Brand strategy. You can't prompt your way to a positioning that resonates. Strategy requires understanding context, constraints, and human behaviour in ways that LLMs don't.

• Code architecture. AI writes functions well but designs systems poorly. It optimises locally, not globally. I've learned not to let it make structural decisions.

• Client communication. Empathy, reading the room, knowing when to push back — these are fundamentally human skills.

My actual toolkit

Here's what I use daily and how:

• Claude (Anthropic): Primary coding assistant. I use Claude Code for scaffolding, debugging, and refactoring. It understands project context in a way that one-shot tools don't.

• Cursor: AI-enhanced editor for larger refactors. The tab completion alone saves hours per week.

• ChatGPT: Content brainstorming and research synthesis. Better for conversational exploration than structured output.

• Midjourney: Concept exploration only. I never ship AI-generated images, but it's useful for mood boards and visual direction.

The uncomfortable truth

AI makes me faster, not better. The quality of my design work hasn't improved because of AI — it's improved because of experience, feedback, and practice. What's changed is the volume of work I can ship as a solo practitioner.

Before AI tools, I couldn't realistically build a full-stack product, write all the content, and handle the design for a client project in a week. Now I can. The leverage is real, but it's leverage on existing skill — not a replacement for it.

What I tell other designers

Learn to code, even if AI writes most of it. The designers who thrive in the next decade will be the ones who can evaluate generated output, not just consume it. You need enough technical literacy to know when the AI is wrong — because it's wrong often, and confidently.

The tools will keep getting better. Your taste, judgment, and ability to understand humans won't be automated. Invest there.